arch/powerpc/kernel/setup_64.c

   1 /*
   2  *
   3  * Common boot and setup code.
   4  *
   5  * Copyright (C) 2001 PPC64 Team, IBM Corp
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #undef DEBUG
  14
  15 #include <linux/module.h>
  16 #include <linux/string.h>
  17 #include <linux/sched.h>
  18 #include <linux/init.h>
  19 #include <linux/kernel.h>
  20 #include <linux/reboot.h>
  21 #include <linux/delay.h>
  22 #include <linux/initrd.h>
  23 #include <linux/ide.h>
  24 #include <linux/seq_file.h>
  25 #include <linux/ioport.h>
  26 #include <linux/console.h>
  27 #include <linux/utsname.h>
  28 #include <linux/tty.h>
  29 #include <linux/root_dev.h>
  30 #include <linux/notifier.h>
  31 #include <linux/cpu.h>
  32 #include <linux/unistd.h>
  33 #include <linux/serial.h>
  34 #include <linux/serial_8250.h>
  35 #include <linux/bootmem.h>
  36 #include <asm/io.h>
  37 #include <asm/kdump.h>
  38 #include <asm/prom.h>
  39 #include <asm/processor.h>
  40 #include <asm/pgtable.h>
  41 #include <asm/smp.h>
  42 #include <asm/elf.h>
  43 #include <asm/machdep.h>
  44 #include <asm/paca.h>
  45 #include <asm/time.h>
  46 #include <asm/cputable.h>
  47 #include <asm/sections.h>
  48 #include <asm/btext.h>
  49 #include <asm/nvram.h>
  50 #include <asm/setup.h>
  51 #include <asm/system.h>
  52 #include <asm/rtas.h>
  53 #include <asm/iommu.h>
  54 #include <asm/serial.h>
  55 #include <asm/cache.h>
  56 #include <asm/page.h>
  57 #include <asm/mmu.h>
  58 #include <asm/lmb.h>
  59 #include <asm/iseries/it_lp_naca.h>
  60 #include <asm/firmware.h>
  61 #include <asm/xmon.h>
  62 #include <asm/udbg.h>
  63 #include <asm/kexec.h>
  64
  65 #include "setup.h"
  66
  67 #ifdef DEBUG
  68 #define DBG(fmt...) udbg_printf(fmt)
  69 #else
  70 #define DBG(fmt...)
  71 #endif
  72
  73 int have_of = 1;
  74 int boot_cpuid = 0;
  75 dev_t boot_dev;
  76 u64 ppc64_pft_size;
  77
  78 /* Pick defaults since we might want to patch instructions
  79  * before we've read this from the device tree.
  80  */
  81 struct ppc64_caches ppc64_caches = {
  82         .dline_size = 0x80,
  83         .log_dline_size = 7,
  84         .iline_size = 0x80,
  85         .log_iline_size = 7
  86 };
  87 EXPORT_SYMBOL_GPL(ppc64_caches);
  88
  89 /*
  90  * These are used in binfmt_elf.c to put aux entries on the stack
  91  * for each elf executable being started.
  92  */
  93 int dcache_bsize;
  94 int icache_bsize;
  95 int ucache_bsize;
  96
  97 #ifdef CONFIG_MAGIC_SYSRQ
  98 unsigned long SYSRQ_KEY;
  99 #endif /* CONFIG_MAGIC_SYSRQ */
 100
 101
 102 #ifdef CONFIG_SMP
 103
 104 static int smt_enabled_cmdline;
 105
 106 /* Look for ibm,smt-enabled OF option */
 107 static void check_smt_enabled(void)
 108 {
 109         struct device_node *dn;
 110         char *smt_option;
 111
 112         /* Allow the command line to overrule the OF option */
 113         if (smt_enabled_cmdline)
 114                 return;
 115
 116         dn = of_find_node_by_path("/options");
 117
 118         if (dn) {
 119                 smt_option = (char *)get_property(dn, "ibm,smt-enabled", NULL);
 120
 121                 if (smt_option) {
 122                         if (!strcmp(smt_option, "on"))
 123                                 smt_enabled_at_boot = 1;
 124                         else if (!strcmp(smt_option, "off"))
 125                                 smt_enabled_at_boot = 0;
 126                 }
 127         }
 128 }
 129
 130 /* Look for smt-enabled= cmdline option */
 131 static int __init early_smt_enabled(char *p)
 132 {
 133         smt_enabled_cmdline = 1;
 134
 135         if (!p)
 136                 return 0;
 137
 138         if (!strcmp(p, "on") || !strcmp(p, "1"))
 139                 smt_enabled_at_boot = 1;
 140         else if (!strcmp(p, "off") || !strcmp(p, "0"))
 141                 smt_enabled_at_boot = 0;
 142
 143         return 0;
 144 }
 145 early_param("smt-enabled", early_smt_enabled);
 146
 147 #else
 148 #define check_smt_enabled()
 149 #endif /* CONFIG_SMP */
 150
 151 /* Put the paca pointer into r13 and SPRG3 */
 152 void __init setup_paca(int cpu)
 153 {
 154         local_paca = &paca[cpu];
 155         mtspr(SPRN_SPRG3, local_paca);
 156 }
 157
 158 /*
 159  * Early initialization entry point. This is called by head.S
 160  * with MMU translation disabled. We rely on the "feature" of
 161  * the CPU that ignores the top 2 bits of the address in real
 162  * mode so we can access kernel globals normally provided we
 163  * only toy with things in the RMO region. From here, we do
 164  * some early parsing of the device-tree to setup out LMB
 165  * data structures, and allocate & initialize the hash table
 166  * and segment tables so we can start running with translation
 167  * enabled.
 168  *
 169  * It is this function which will call the probe() callback of
 170  * the various platform types and copy the matching one to the
 171  * global ppc_md structure. Your platform can eventually do
 172  * some very early initializations from the probe() routine, but
 173  * this is not recommended, be very careful as, for example, the
 174  * device-tree is not accessible via normal means at this point.
 175  */
 176
 177 void __init early_setup(unsigned long dt_ptr)
 178 {
 179         /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
 180         setup_paca(0);
 181
 182         /* Enable early debugging if any specified (see udbg.h) */
 183         udbg_early_init();
 184
 185         DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
 186
 187         /*
 188          * Do early initializations using the flattened device
 189          * tree, like retreiving the physical memory map or
 190          * calculating/retreiving the hash table size
 191          */
 192         early_init_devtree(__va(dt_ptr));
 193
 194         /* Now we know the logical id of our boot cpu, setup the paca. */
 195         setup_paca(boot_cpuid);
 196
 197         /* Fix up paca fields required for the boot cpu */
 198         get_paca()->cpu_start = 1;
 199         get_paca()->stab_real = __pa((u64)&initial_stab);
 200         get_paca()->stab_addr = (u64)&initial_stab;
 201
 202         /* Probe the machine type */
 203         probe_machine();
 204
 205         setup_kdump_trampoline();
 206
 207         DBG("Found, Initializing memory management...\n");
 208
 209         /*
 210          * Initialize the MMU Hash table and create the linear mapping
 211          * of memory. Has to be done before stab/slb initialization as
 212          * this is currently where the page size encoding is obtained
 213          */
 214         htab_initialize();
 215
 216         /*
 217          * Initialize stab / SLB management except on iSeries
 218          */
 219         if (cpu_has_feature(CPU_FTR_SLB))
 220                 slb_initialize();
 221         else if (!firmware_has_feature(FW_FEATURE_ISERIES))
 222                 stab_initialize(get_paca()->stab_real);
 223
 224         DBG(" <- early_setup()\n");
 225 }
 226
 227 #ifdef CONFIG_SMP
 228 void early_setup_secondary(void)
 229 {
 230         struct paca_struct *lpaca = get_paca();
 231
 232         /* Mark enabled in PACA */
 233         lpaca->proc_enabled = 0;
 234
 235         /* Initialize hash table for that CPU */
 236         htab_initialize_secondary();
 237
 238         /* Initialize STAB/SLB. We use a virtual address as it works
 239          * in real mode on pSeries and we want a virutal address on
 240          * iSeries anyway
 241          */
 242         if (cpu_has_feature(CPU_FTR_SLB))
 243                 slb_initialize();
 244         else
 245                 stab_initialize(lpaca->stab_addr);
 246 }
 247
 248 #endif /* CONFIG_SMP */
 249
 250 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
 251 void smp_release_cpus(void)
 252 {
 253         extern unsigned long __secondary_hold_spinloop;
 254         unsigned long *ptr;
 255
 256         DBG(" -> smp_release_cpus()\n");
 257
 258         /* All secondary cpus are spinning on a common spinloop, release them
 259          * all now so they can start to spin on their individual paca
 260          * spinloops. For non SMP kernels, the secondary cpus never get out
 261          * of the common spinloop.
 262          * This is useless but harmless on iSeries, secondaries are already
 263          * waiting on their paca spinloops. */
 264
 265         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
 266                         - PHYSICAL_START);
 267         *ptr = 1;
 268         mb();
 269
 270         DBG(" <- smp_release_cpus()\n");
 271 }
 272 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 273
 274 /*
 275  * Initialize some remaining members of the ppc64_caches and systemcfg
 276  * structures
 277  * (at least until we get rid of them completely). This is mostly some
 278  * cache informations about the CPU that will be used by cache flush
 279  * routines and/or provided to userland
 280  */
 281 static void __init initialize_cache_info(void)
 282 {
 283         struct device_node *np;
 284         unsigned long num_cpus = 0;
 285
 286         DBG(" -> initialize_cache_info()\n");
 287
 288         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
 289                 num_cpus += 1;
 290
 291                 /* We're assuming *all* of the CPUs have the same
 292                  * d-cache and i-cache sizes... -Peter
 293                  */
 294
 295                 if ( num_cpus == 1 ) {
 296                         u32 *sizep, *lsizep;
 297                         u32 size, lsize;
 298                         const char *dc, *ic;
 299
 300                         /* Then read cache informations */
 301                         if (machine_is(powermac)) {
 302                                 dc = "d-cache-block-size";
 303                                 ic = "i-cache-block-size";
 304                         } else {
 305                                 dc = "d-cache-line-size";
 306                                 ic = "i-cache-line-size";
 307                         }
 308
 309                         size = 0;
 310                         lsize = cur_cpu_spec->dcache_bsize;
 311                         sizep = (u32 *)get_property(np, "d-cache-size", NULL);
 312                         if (sizep != NULL)
 313                                 size = *sizep;
 314                         lsizep = (u32 *) get_property(np, dc, NULL);
 315                         if (lsizep != NULL)
 316                                 lsize = *lsizep;
 317                         if (sizep == 0 || lsizep == 0)
 318                                 DBG("Argh, can't find dcache properties ! "
 319                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 320
 321                         ppc64_caches.dsize = size;
 322                         ppc64_caches.dline_size = lsize;
 323                         ppc64_caches.log_dline_size = __ilog2(lsize);
 324                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 325
 326                         size = 0;
 327                         lsize = cur_cpu_spec->icache_bsize;
 328                         sizep = (u32 *)get_property(np, "i-cache-size", NULL);
 329                         if (sizep != NULL)
 330                                 size = *sizep;
 331                         lsizep = (u32 *)get_property(np, ic, NULL);
 332                         if (lsizep != NULL)
 333                                 lsize = *lsizep;
 334                         if (sizep == 0 || lsizep == 0)
 335                                 DBG("Argh, can't find icache properties ! "
 336                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 337
 338                         ppc64_caches.isize = size;
 339                         ppc64_caches.iline_size = lsize;
 340                         ppc64_caches.log_iline_size = __ilog2(lsize);
 341                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 342                 }
 343         }
 344
 345         DBG(" <- initialize_cache_info()\n");
 346 }
 347
 348
 349 /*
 350  * Do some initial setup of the system.  The parameters are those which
 351  * were passed in from the bootloader.
 352  */
 353 void __init setup_system(void)
 354 {
 355         DBG(" -> setup_system()\n");
 356
 357         /*
 358          * Unflatten the device-tree passed by prom_init or kexec
 359          */
 360         unflatten_device_tree();
 361
 362         /*
 363          * Fill the ppc64_caches & systemcfg structures with informations
 364          * retrieved from the device-tree.
 365          */
 366         initialize_cache_info();
 367
 368         /*
 369          * Initialize irq remapping subsystem
 370          */
 371         irq_early_init();
 372
 373 #ifdef CONFIG_PPC_RTAS
 374         /*
 375          * Initialize RTAS if available
 376          */
 377         rtas_initialize();
 378 #endif /* CONFIG_PPC_RTAS */
 379
 380         /*
 381          * Check if we have an initrd provided via the device-tree
 382          */
 383         check_for_initrd();
 384
 385         /*
 386          * Do some platform specific early initializations, that includes
 387          * setting up the hash table pointers. It also sets up some interrupt-mapping
 388          * related options that will be used by finish_device_tree()
 389          */
 390         ppc_md.init_early();
 391
 392         /*
 393          * We can discover serial ports now since the above did setup the
 394          * hash table management for us, thus ioremap works. We do that early
 395          * so that further code can be debugged
 396          */
 397         find_legacy_serial_ports();
 398
 399         /*
 400          * Initialize xmon
 401          */
 402 #ifdef CONFIG_XMON_DEFAULT
 403         xmon_init(1);
 404 #endif
 405         /*
 406          * Register early console
 407          */
 408         register_early_udbg_console();
 409
 410         if (do_early_xmon)
 411                 debugger(NULL);
 412
 413         check_smt_enabled();
 414         smp_setup_cpu_maps();
 415
 416 #ifdef CONFIG_SMP
 417         /* Release secondary cpus out of their spinloops at 0x60 now that
 418          * we can map physical -> logical CPU ids
 419          */
 420         smp_release_cpus();
 421 #endif
 422
 423         printk("Starting Linux PPC64 %s\n", system_utsname.version);
 424
 425         printk("-----------------------------------------------------\n");
 426         printk("ppc64_pft_size                = 0x%lx\n", ppc64_pft_size);
 427         printk("physicalMemorySize            = 0x%lx\n", lmb_phys_mem_size());
 428         printk("ppc64_caches.dcache_line_size = 0x%x\n",
 429                ppc64_caches.dline_size);
 430         printk("ppc64_caches.icache_line_size = 0x%x\n",
 431                ppc64_caches.iline_size);
 432         printk("htab_address                  = 0x%p\n", htab_address);
 433         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 434 #if PHYSICAL_START > 0
 435         printk("physical_start                = 0x%x\n", PHYSICAL_START);
 436 #endif
 437         printk("-----------------------------------------------------\n");
 438
 439         DBG(" <- setup_system()\n");
 440 }
 441
 442 #ifdef CONFIG_IRQSTACKS
 443 static void __init irqstack_early_init(void)
 444 {
 445         unsigned int i;
 446
 447         /*
 448          * interrupt stacks must be under 256MB, we cannot afford to take
 449          * SLB misses on them.
 450          */
 451         for_each_possible_cpu(i) {
 452                 softirq_ctx[i] = (struct thread_info *)
 453                         __va(lmb_alloc_base(THREAD_SIZE,
 454                                             THREAD_SIZE, 0x10000000));
 455                 hardirq_ctx[i] = (struct thread_info *)
 456                         __va(lmb_alloc_base(THREAD_SIZE,
 457                                             THREAD_SIZE, 0x10000000));
 458         }
 459 }
 460 #else
 461 #define irqstack_early_init()
 462 #endif
 463
 464 /*
 465  * Stack space used when we detect a bad kernel stack pointer, and
 466  * early in SMP boots before relocation is enabled.
 467  */
 468 static void __init emergency_stack_init(void)
 469 {
 470         unsigned long limit;
 471         unsigned int i;
 472
 473         /*
 474          * Emergency stacks must be under 256MB, we cannot afford to take
 475          * SLB misses on them. The ABI also requires them to be 128-byte
 476          * aligned.
 477          *
 478          * Since we use these as temporary stacks during secondary CPU
 479          * bringup, we need to get at them in real mode. This means they
 480          * must also be within the RMO region.
 481          */
 482         limit = min(0x10000000UL, lmb.rmo_size);
 483
 484         for_each_possible_cpu(i)
 485                 paca[i].emergency_sp =
 486                 __va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
 487 }
 488
 489 /*
 490  * Called into from start_kernel, after lock_kernel has been called.
 491  * Initializes bootmem, which is unsed to manage page allocation until
 492  * mem_init is called.
 493  */
 494 void __init setup_arch(char **cmdline_p)
 495 {
 496         ppc64_boot_msg(0x12, "Setup Arch");
 497
 498         *cmdline_p = cmd_line;
 499
 500         /*
 501          * Set cache line size based on type of cpu as a default.
 502          * Systems with OF can look in the properties on the cpu node(s)
 503          * for a possibly more accurate value.
 504          */
 505         dcache_bsize = ppc64_caches.dline_size;
 506         icache_bsize = ppc64_caches.iline_size;
 507
 508         /* reboot on panic */
 509         panic_timeout = 180;
 510
 511         if (ppc_md.panic)
 512                 setup_panic();
 513
 514         init_mm.start_code = PAGE_OFFSET;
 515         init_mm.end_code = (unsigned long) _etext;
 516         init_mm.end_data = (unsigned long) _edata;
 517         init_mm.brk = klimit;
 518
 519         irqstack_early_init();
 520         emergency_stack_init();
 521
 522         stabs_alloc();
 523
 524         /* set up the bootmem stuff with available memory */
 525         do_init_bootmem();
 526         sparse_init();
 527
 528 #ifdef CONFIG_DUMMY_CONSOLE
 529         conswitchp = &dummy_con;
 530 #endif
 531
 532         ppc_md.setup_arch();
 533
 534         paging_init();
 535         ppc64_boot_msg(0x15, "Setup Done");
 536 }
 537
 538
 539 /* ToDo: do something useful if ppc_md is not yet setup. */
 540 #define PPC64_LINUX_FUNCTION 0x0f000000
 541 #define PPC64_IPL_MESSAGE 0xc0000000
 542 #define PPC64_TERM_MESSAGE 0xb0000000
 543
 544 static void ppc64_do_msg(unsigned int src, const char *msg)
 545 {
 546         if (ppc_md.progress) {
 547                 char buf[128];
 548
 549                 sprintf(buf, "%08X\n", src);
 550                 ppc_md.progress(buf, 0);
 551                 snprintf(buf, 128, "%s", msg);
 552                 ppc_md.progress(buf, 0);
 553         }
 554 }
 555
 556 /* Print a boot progress message. */
 557 void ppc64_boot_msg(unsigned int src, const char *msg)
 558 {
 559         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 560         printk("[boot]%04x %s\n", src, msg);
 561 }
 562
 563 /* Print a termination message (print only -- does not stop the kernel) */
 564 void ppc64_terminate_msg(unsigned int src, const char *msg)
 565 {
 566         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
 567         printk("[terminate]%04x %s\n", src, msg);
 568 }
 569
 570 void cpu_die(void)
 571 {
 572         if (ppc_md.cpu_die)
 573                 ppc_md.cpu_die();
 574 }
 575
 576 #ifdef CONFIG_SMP
 577 void __init setup_per_cpu_areas(void)
 578 {
 579         int i;
 580         unsigned long size;
 581         char *ptr;
 582
 583         /* Copy section for each CPU (we discard the original) */
 584         size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
 585 #ifdef CONFIG_MODULES
 586         if (size < PERCPU_ENOUGH_ROOM)
 587                 size = PERCPU_ENOUGH_ROOM;
 588 #endif
 589
 590         for_each_possible_cpu(i) {
 591                 ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
 592                 if (!ptr)
 593                         panic("Cannot allocate cpu data for CPU %d\n", i);
 594
 595                 paca[i].data_offset = ptr - __per_cpu_start;
 596                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 597         }
 598 }
 599 #endif